BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Vienna
BEGIN:DAYLIGHT
DTSTART:20220327T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20211031T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260405T153415Z
UID:1648131300@ist.ac.at
DTSTART:20220324T151500
DTEND:20220324T170000
DESCRIPTION:Speaker: Felix Peterson\nhosted by Christoph Lampert\nAbstract:
  Classic algorithms and machine learning systems like neural networksare b
 oth abundant in everyday life. While classic computerscience algorithms ar
 e suitable for precise execution of exactly definedtasks such as finding t
 he shortest path in a large graph\, neuralnetworks allow learning from dat
 a to predict the most likely answer inmore complex tasks such as image cla
 ssification\, which cannot bereduced to an exact algorithm. In the talk\, 
 we explore combiningboth concepts leading to more robust\, better performi
 ng\, more interpretable\,more computationally efficient\, and most importa
 ntly dataefficient architectures. Using algorithmic supervision a neural n
 etwork can learnfrom or in conjunction with an algorithm. When integrating
  an algorithminto a neural architecture\, it is important that the algorit
 hm isdifferentiable such that the architecture can be trained end-to-end a
 ndgradients can be propagated back through the algorithm in a meaningfulwa
 y. To make algorithms differentiable\, I discuss a generalmethod for conti
 nuously relaxing algorithms by perturbing variableswith logistic distribut
 ions. In addition\, I discuss specialized differentiablealgorithms such as
  differentiable sorting networks\, and efficientand effective differentiab
 le sorting and ranking operators allowingsorting and ranking supervision. 
 Furthermore\, I delve into differentiablerendering\, specifically\, the ge
 neralized differentiable renderer GenDR.
LOCATION:Join Zoom Meeting https://istaustria.zoom.us/j/69657308623?pwd=MTd
 LS1I4T084cU43OEs3TFN2NjNHdz09  Meeting ID: 696 5730 8623 Passcode: 098438\
 , ISTA
ORGANIZER:
SUMMARY:Felix Peterson: Integrating Algorithms into Neural Networks
URL:https://talks-calendar.ista.ac.at/events/3590
END:VEVENT
END:VCALENDAR